Beyond simple interaction: Uncovering the perception-interaction intrinsic mechanism of generative AI agents—A multi-modal big data analysis with PLS-SEM and fsQCA

Abstract

Generative Artificial Intelligence (GenAI) is increasingly being adopted across industries, yet existing literature has not fully explored the unique traits and the complex mechanism it introduces. To address this gap, this study investigates the unique characteristics of GenAI agents and their impact on user interaction behaviors. By analyzing user-generated text and AI-generated images from the Character.AI platform, we examine three key perceptual characteristics: social personalization, functional customization, and emotional affordance. Through multi-modal machine learning approaches combining Structural Topic Modeling (STM) and Facial Action Coding System (FACS), we propose the “perceived characteristics of GenAI agent-empathy-interactive willingness” (PCoGenAI-E-IW) theoretical model to explore how user perceptions transform into interactive behaviors. Furthermore, the PLS-SEM analysis and configurational approach identify 10 distinct variable combinations that influence users’ interaction willingness. The findings validate our multi-modal analytical framework while providing valuable empirical evidence for marketing strategy formulation, service experience optimization, and theoretical advancement in human-AI interaction research.